As we continue to innovate and improve Nearmap AI, we mark major changes in performance and capability with a new "generation" of content. This section provides a description of the new performance, features and content introduced with each generation of Nearmap AI data. Multiple generations will be available for export at any given time, with the latest generations typically on the most recent capture dates.
AI generations currently available:
A table mapping each generation to the various access mechanisms can be found at the bottom of the Access AI Content page.
How Versions Work
We have introduced a new versioning systems, to help customers keep track of improvements over time. It is loosely based on the Semantic Versioning Specification (SemVer) used for software development, but we have adapted it to our needs with AI generated geospatial data. It is very important to note that rather than versioning a piece of software, this represents the a combined version of the system that produces the data, and the data itself. Multiple versions and generations of data may be available at any given point, and from Gen 3 onwards, the AI System Version will always be denoted strictly by a version of the format shown above.An "AI System Version" (or
systemVersion in the AI Feature API responses) looks like this:
The version is comprised of three components, separated by hyphens: The generation, the config name, and the minor/patch versions.
The generation is a numerically ordered set of improvements that represent major enhancements to our offering. Usually, they will involve a new raster model, but in the case of Gen 2 to Gen 3, it involved a complete rework of the vectorisation system.
A good rule of thumb for a generation change is that it breaks backwards compatibility (as per the "major" version in SemVer). This means that performing change detection between two dates must be done with great care, as the improvements in the more recent generation may be more prevalent than the genuine on-the-ground changes when looking for differences. Other aspects of the returned data format may change and break compatibility.
The config name is a pair of words (usually weather themed!) separated by an underscore that denote a particular configuration of that generation. For Gen 3, there is only one formal Config Name (
winter_sunrise) for our standard production data, as described in the relevant AI Pack pages. If custom data is provided in an offline delivery, the config name will always be "
Minor / Patch Versions
Minor Version (Y.z)
The Minor Version may be incremented if new attributes or improvements are introduced on an existing generation and config name. The intent is that this is backwards compatible - most attributes will remain unchanged (in fact, if the addition is a new AI Pack, customers of existing AI Packs may not notice any difference at all).
Patch Version (y.Z)
The patch version is reserved for changes to the data that do not change format, or add new attributes. One example might be if a problem with a corner case of the data in some location is discovered, it will be fixed, and the version incremented. Due to the scale of the data we work with (billions of polygons and their metadata), most customers will not noticed any difference at all when the patch version is increased. Note that rather than manually fixing an error, we always improve the "factory" (the code that produces the data), and re-run any areas that we deem to have been impacted.